Forced 3 frames on compute buffer readback regardless of how long those frames are.

EDIT: Bug Report Case: 1304040

When running even the simplest ComputeShader and then reading the result back with AsyncReadback it seems obvious the readback is forced to wait 2 or 3 frames.
What's really weird about it is that this happens regardless of whether V-blank is ON or not. In other words with v-blank OFF readback "only" waits about 3 * 3ms = 9ms but with v-blank set to "every other" it waits 3 * 33ms = 100ms.
I hope it's plain to see this makes Compute unnecessarily cumbersome to use in low FPS scenarios.

I didn't report this as a bug yet because I'm not sure if it's not forced by the driver / architecture somehow.

Unity 2018.2 as well as 2020.2
Windows 10 x64
Direct 3D 11

example code:

Note: Command Buffer is used!

ComputeReadbackTest.cs

Code (CSharp):

1.  
2. using System;
3. using System.Collections.Generic;
4. using System.Threading;
5.  
6. using UnityEngine;
7. using UnityEngine.Rendering;
8.  
9.  
10. public class ComputeReadbackTest : MonoBehaviour
11. {
12.   private ComputeShader testShader;
13.   private int CSMainKernelIndex;
14.   private int resultBuffID;
15.   private ComputeBuffer resultCB;
16.   private int frame;
17.   private volatile bool nextDispatchOK = true;
18.  
19.  
20.   private void Start()
21.   {
22.     testShader = Resources.Load<ComputeShader>("TestComputeShader");
23.     CSMainKernelIndex = testShader.FindKernel("CSMain");
24.     resultBuffID = Shader.PropertyToID("Result");
25.     resultCB = new ComputeBuffer(1, 16, ComputeBufferType.Default);
26.     testShader.SetBuffer(CSMainKernelIndex, resultBuffID, resultCB);
27.   }
28.  
29.   private void Update()
30.   {
31.     if(!nextDispatchOK)
32.       return;
33.  
34.     nextDispatchOK = false;
35.     CommandBuffer commandBuffer = new CommandBuffer();
36.     commandBuffer.DispatchCompute(testShader, CSMainKernelIndex, 1, 1, 1);
37.     frame = Time.frameCount;
38.  
39.     // 2018.2 or newer
40.     commandBuffer.RequestAsyncReadback(resultCB, AsyncGPUReadbackCallback);
41.  
42.     Graphics.ExecuteCommandBuffer(commandBuffer);
43.  
44.     if(nextDispatchOK)
45.       Debug.Log("Callback called.");
46.   }
47.  
48.   private void AsyncGPUReadbackCallback(AsyncGPUReadbackRequest req)
49.   {
50.     nextDispatchOK = true;
51.  
52.     if(req.hasError)
53.       Debug.LogError("Readback error!");
54.  
55.     if(frame == Time.frameCount)
56.       Debug.Log("Same frame!");
57.     else
58.       Debug.Log("start frame: " + frame + "; end frame: " + Time.frameCount);
59.   }
60. }
61.  

TestComputeShader.compute

Code (CSharp):

1.  
2. #pragma kernel CSMain
3.  
4. RWStructuredBuffer<float4> Result;
5.  
6. [numthreads(1, 1, 1)]
7. void CSMain(uint3 id : SV_DispatchThreadID)
8. {
9.     Result[id.x] = float4(id.x & id.y, (id.x & 15) / 15.0, (id.y & 15) / 15.0, 0.0);
10. }
11.  

 

Issue doesn't seem to raise any interest on the forum so I decided to report it. Hopefully more luck in the issue tracker.